> ## Documentation Index
> Fetch the complete documentation index at: https://docs.unleashx.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Create Evaluation Criteria

> Create a criteria mapped to an evaluation config.

Create a criteria mapped to an evaluation config. `eval_id` is required.

Updating an existing one: see [Update Evaluation Criteria](https://docs.unleashx.ai/api-reference/evaluation/update-criteria.md).

## Scoring fields

Each call gets a score from 0 to 100 for this criteria. The thresholds decide the result:

| Score | Result |
| - | - |
| `pass_above` or more | Pass |
| Between `fail_below` and `pass_above` | Review |
| Below `fail_below` | Fail |

With the UI defaults (`pass_above` 70, `fail_below` 40) a score of 85 passes, 55 is "Review" and 30 fails. `review_below` is the top of the Review band, so it should be `pass_above - 1` (69 here). The UI sets it for you, and you can leave it out.

| Field | Meaning | UI default |
| - | - | - |
| `accuracy_percentage` | Weight of this criteria in the overall score. It is relative to the other criteria of the same evaluation (weights 10, 10, 20 give 25%, 25%, 50%). Minimum 1. | 10 |
| `pass_above` | Pass at or above this score (0-100) | 70 |
| `fail_below` | Fail below this score (0-100) | 40 |
| `review_below` | Top of the Review band, normally `pass_above - 1` | 69 |
| `scoring_rubric` | Optional text that explains how to score. Saved with the criteria; the UI does not set it. | none |

Keep `fail_below` lower than `pass_above`.

## Criteria

Pick `criteria` from this list (send the name exactly). If you do not send `evaluation_prompt` when creating, the default prompt below is used.

| Criteria | Default prompt |
| - | - |
| `Empathy` | Evaluate the agent's ability to recognize and validate customer emotions while maintaining professionalism. Score based on authentic acknowledgment of concerns and compassionate communication. |
| `Language Switch` | Assess the agent's linguistic proficiency in seamlessly transitioning between English and Hindi. Rate fluency, grammatical accuracy, and cultural appropriateness throughout the conversation. |
| `Turn Taking` | Analyze conversation balance by measuring speaking time distribution and interruption patterns. Score higher when both participants have equitable opportunities to communicate. |
| `Context Awareness` | Evaluate whether the agent references relevant customer history and prior interactions. Penalize redundant information requests and score based on personalized, informed responses. |
| `Recovery From Errors` | Assess the agent's response to mistakes through accountability, explanation, and remediation. Score based on professional composure and customer confidence restoration. |
| `Intent Recognition` | Evaluate the agent's ability to identify the underlying customer need beyond surface requests. Score based on clarifying questions and accurate problem diagnosis. |
| `Greeting Accuracy` | Assess the professionalism and completeness of the call opening, including identity verification and call objective clarity. Score based on adherence to service standards. |
| `Intent Confirmation` | Evaluate the agent's use of summarization and clarification techniques before proceeding. Score based on explicit customer acknowledgment of understanding. |
| `Probing Questions` | Assess the strategic depth and relevance of questioning. Score based on open-ended inquiry, follow-up rigor, and absence of leading or manipulative framing. |
| `Call Closing` | Evaluate closure completeness including action item recap, next-step clarity, and satisfaction verification. Score based on professional sign-off and unresolved issue identification. |
| `No Offensive Content` | Conduct thorough language review for profanity, disrespect, or cultural insensitivity. High scores require zero violations of professional communication standards. |
| `No Misrepresentation` | Verify factual accuracy, proper disclosure of limitations, and absence of exaggerated claims. Score based on compliance with truthfulness and transparency standards. |
| `Privacy Compliance` | Assess protection of personally identifiable information and adherence to regulatory standards. Score based on identity verification protocols and confidential information handling. |
| `Confidence Level` | Evaluate the agent's demonstration of product knowledge and solution competence. Score based on conviction in recommendations balanced with appropriate acknowledgment of limitations. |
| `Input Metadata Consistency` | Verify alignment between transcript content, call metadata, and recorded agent information. Score based on absence of contradictions across all documentation fields. |


## OpenAPI

````yaml api-reference/openapi.json POST /evaluation/criteria/create-update/
openapi: 3.1.0
info:
  title: >-
    UnleashX - Build human like conversations | Voice Agents | Automations | AI
    Workforce
  version: 1.0.0
  description: UnleashX - Your home for human like conversations
servers:
  - url: https://www.tryunleashx.com/api/v1/global/
security:
  - bearerAuth: []
paths:
  /evaluation/criteria/create-update/:
    post:
      tags:
        - Evaluation
      summary: Create evaluation criteria
      description: Create a criteria mapped to an evaluation config.
      parameters:
        - name: token
          in: header
          required: true
          description: API token for authentication
          schema:
            type: string
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              required:
                - eval_id
              properties:
                eval_id:
                  type: integer
                  description: Evaluation ID
                criteria:
                  type: string
                  enum:
                    - Empathy
                    - Language Switch
                    - Turn Taking
                    - Context Awareness
                    - Recovery From Errors
                    - Intent Recognition
                    - Greeting Accuracy
                    - Intent Confirmation
                    - Probing Questions
                    - Call Closing
                    - No Offensive Content
                    - No Misrepresentation
                    - Privacy Compliance
                    - Confidence Level
                    - Input Metadata Consistency
                  description: >-
                    Criteria to score, picked from the list. When creating and
                    `evaluation_prompt` is not sent, the default prompt of this
                    criteria is used.
                evaluation_prompt:
                  type: string
                  description: >-
                    Prompt for this criteria. Optional when creating: the
                    default prompt of the chosen criteria is used.
                scoring_rubric:
                  type: string
                  description: >-
                    Optional free text that explains how to score this criteria
                    (for example what a 90+ vs a 50 looks like). It is saved
                    with the criteria. The in-app UI does not set it, so you can
                    leave it out.
                pass_above:
                  type: integer
                  description: >-
                    Score (0-100) at or above which the call passes this
                    criteria (UI: "Pass ≥"). UI default 70.
                  minimum: 0
                  maximum: 100
                review_below:
                  type: integer
                  description: >-
                    Highest score (0-100) that still counts as "Review". The
                    Review band is the scores between fail_below and pass_above.
                    The UI sets it automatically to pass_above - 1, so you can
                    usually leave it out. UI default 69.
                  minimum: 0
                  maximum: 100
                fail_below:
                  type: integer
                  description: >-
                    Score (0-100) below which the call fails this criteria (UI:
                    "Fail <"). UI default 40.
                  minimum: 0
                  maximum: 100
                accuracy_percentage:
                  type: integer
                  description: >-
                    Weight of this criteria in the overall score (UI: "Weight").
                    Minimum 1; UI default 10. It is relative: with weights 10,
                    10 and 20 the criteria count for 25%, 25% and 50% of the
                    overall score.
                  minimum: 1
      responses:
        '200':
          description: Success
          content:
            application/json:
              schema:
                type: object
      security: []
components:
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer

````

This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.